Analytics in banking:

Time to realize the value

Banking April 2017

Amit Garg
Davide Grande
Gloria Macias-Lizaso Miranda
Christoph Sporleder
Eckart Windhagen
 Analytics in banking: Time to realize
the value
By establishing analytics as a true business discipline, banks can grasp the
enormous potential.

Consider three recent examples of the power of Results like these are the good news about
analytics in banking: analytics. But they are also the bad news. While
many such projects generate eye-popping returns
ƒƒ To counter a shrinking customer base, a on investment, banks find it difficult to scale
European bank tried a number of retention them up; the financial impact from even several
techniques focusing on inactive customers, great analytics efforts is often insignificant
but without significant results. Then it turned for the enterprise P&L. Some executives are
to machine-learning algorithms that predict even concluding that while analytics may be
which currently active customers are likely to a welcome addition to certain activities, the
reduce their business with the bank. This new difficulties in scaling it up mean that, at best,
understanding gave rise to a targeted campaign it will be only a sideline to the traditional
that reduced churn by 15 percent. businesses of financing, investments, and
transactions and payments.
ƒƒ A US bank used machine learning to study the
discounts its private bankers were offering to In our view, that’s shortsighted. Analytics can
customers. Bankers claimed that they offered involve much more than just a set of discrete
them only to valuable ones and more than made projects. If banks put their considerable strategic
up for them with other, high-margin business. and organizational muscle into analytics, it can
The analytics showed something different: and should become a true business discipline.
patterns of unnecessary discounts that could Business leaders today may only faintly remember
easily be corrected. After the unit adopted the what banking was like before marketing and
changes, revenues rose by 8 percent within a sales, for example, became a business discipline,
few months. sometime in the 1970s. They can more easily recall
the days when information technology was just six
ƒƒ A top consumer bank in Asia enjoyed a large guys in the basement with an IBM mainframe. A
market share but lagged behind its competitors look around banks today—at all the businesses and
in products per customer. It used advanced processes powered by extraordinary IT—is a strong
analytics to explore several sets of big data: reminder of the way a new discipline can radically
customer demographics and key characteristics, reshape the old patterns of work. Analytics has
products held, credit-card statements, that potential.
transaction and point-of-sale data, online and
mobile transfers and payments, and credit- Why? Three factors are coming together to kick
bureau data. The bank discovered unsuspected off the coming heyday. First, consider advances
similarities that allowed it to define 15,000 in technologies. The availability of information
microsegments in its customer base. It then built is booming: in the past few years, the amount
a next-product-to-buy model that increased the of meaningful data—true signal, not noise—has
likelihood to buy three times over. grown exponentially, while the size and cost

2
of processors decreased. By 2020, about now carry them forward into actions that can
1.7 megabytes a second of new information will drive meaningful change. The canvas is as broad
be created for every human being on the planet. as a bank itself. Rich real-time data—numbers,
Businesses have opened their minds, freely yes, but also text, voice, and images—now exist
adapting new analytical techniques that in the past for literally every action that customers make,
might have been dismissed as too impractical and every product that banks sell, and every process
theoretical for the real world. that banks use to deliver those products. In this
article, we will explore the vast opportunities, as
And those techniques have improved radically. We well as the problems of integration and scaling
are well past simple linear regressions—machine that keep banks from making analytics a coherent
learning now features support vector machines, discipline. We will then suggest the strategic and
random forests, gradient boosting, and many other organizational elements that banks need to realize
astonishing algorithms. Any company’s ability the analytics dream.
to perform these analytics has been significantly
boosted by the exponential increase of computing Ready for prime time
power (which makes it possible to undertake, in As the saying goes, “The future is already here. It’s
just seconds, an analysis that in the past would just not evenly distributed.” Banks—and companies
have taken weeks) and by new data-storage in every other industry—are already deploying
technologies, such as Hadoop. advanced analytics to move their businesses
forward. We see three ways it can generate a
Second, banks in many regions are under meaningful increase in profits (Exhibit 1).
enormous economic pressure. Our latest research
finds that of the top 500 institutions around the ƒƒ Advanced analytics can help banks wring small
world, 54 percent are priced below book value. In improvements out of almost all their everyday
2014 we calculated that just 18 percent of banks activities, boosting the traditional P&L levers.
captured all the value in the industry. Recognizing Potential moves include the following:
this reality, banks have tried all manner of
improvements, especially digitization and cost —— Accelerating growth, even in an anemic
cutting. But these moves have taken them only so environment. Deeper and more detailed
far; something new is needed. profiles of customers, together with
transactional and trading analytics, can
Those digitization efforts underlie the third factor improve the acquisition and retention of
pushing analytics. Much of a typical bank is now clients, as well as cross- and upselling.
digitized and throwing off data by the terabyte. A For example, one bank used credit-card
mostly manual bank would have serious difficulty transactional data (from both its own
using advanced analytics; at digital banks, the terminals and those of other banks)
highways are already paved. to develop offers that gave customers
incentives to make regular purchases from
Put it all together, and you get advanced analytics: one of the bank’s merchants. This boosted
industrial-scale solutions to exploit data for the bank’s commissions, added revenue for
authentic business insights and vastly improved its merchants, and provided more value to
decision making. The tools are there; banks must the customer.

3
 CDP 2016
Analytics in banking Time to realize the value
Exhibit 1 of 2

Exhibit 1 Three ways advanced analytics can generate an increase in profits.

Boosting traditional P&L levers

• Accelerating growth
• Enhancing productivity
• Improving risk control Finding new
sources
of growth Delivering
the digital
bank

—— Enhancing productivity. Every banking economic returns from their considerable

process can become faster and more investments. We estimate that G-SIBs
effective. Among other things, banks can can take out up to $1 billion a year in costs
use advanced analytics to provide faster through a simplified portfolio of data
and more accurate responses to regulatory repositories—building on work that most
requests and give teams analytics-enhanced banks have already done—and through
decision support. One bank we know new analytics that produce more accurate
used machine learning to understand the regulatory reports and deliver them more
way the characteristics of code affected a quickly. D-SIBs can save up to $400 million
mainframe’s running time and the resulting annually. Further out, banks will be able to
costs; by optimizing the code, it cut them use analytics to reduce fraud losses.
by 15 percent. Another bank used new
algorithms to predict the cash required ƒƒ A second vector of impact is the way that
at each of its ATMs across the country, analytics can help deliver the promise of
combining this with route-optimization digital banks and offer a much better customer
techniques to save money. experience at a fraction of the current cost. In
some regions up to 65 percent of customers
—— I mproving risk control. Banks can lower now interact with their banks via multiple
their risk costs through analytics- channels. Their paths through them are
aided techniques, such as digital credit extraordinarily complex: they often start in
assessment, advanced early-warning one channel, perform intermediate steps in
systems, next-generation stress testing, and others, and finish in yet another—with plenty
credit-collection analytics. The expense of of pauses and information-gathering loops
compliance and control has soared in recent along the way. Successful digital banks deliver
years, and banks can use analytics to get a truly seamless multichannel experience by

4
 gathering real-time data and using analytics to with advanced-analytics techniques (mostly
understand the customer and build the proper through specialized teams loosely connected to the
(and always consistent) journey view. corporate center). But the expected results have not
arrived. A few banks have yet to begin.
ƒƒ Finally, analytics can help banks find new
sources of growth, and even new business The started-but-stuck ones are running into a
models. Banks may be able to reap income from number of problems. At the highest level, their
their data—for example, by sharing customer- efforts remain unconnected and subscale; they
analytics capabilities with new ecosystem have not yet tied together their disparate efforts
partners, such as telecom companies or into a single, unified business discipline. Tactically,
retailers. Taken to a logical but not implausible we see banks making unforced errors such as these:
extreme, banks can use data and analytics to
shape a new business model and out-fintech ƒƒ not quantifying the potential of analytics at a
the fintechs. The bank as data company can sit detailed level
at the center of a consumer ecosystem where
the revenue pools include not just banking but ƒƒ not engaging business leaders early and to
also many other B2C and B2B businesses. Great develop models that really solve their problems
analytics isn’t the only requirement here: banks and that they trust and will use—not a “black box”
must get many other things right to be relevant
to and trusted by customers. But that can be ƒƒ falling into the “pilot trap”: continually trying
done, and already more than a dozen leading new experiments but not following through by
banks are taking positive steps in this direction. fully industrializing and adopting them

Not without pitfalls ƒƒ investing too much up front in data

In our recent survey,1 we found that almost every infrastructure and data quality, without a clear
bank lists advanced analytics among its top five view of the planned use or the expected returns
priorities. Most plan to invest further in these
techniques. A few banks are already seeing the ƒƒ not seeking cooperation from businesses that
rewards. These leaders have built substantial protect rather than share their data
foundations by establishing data lakes and
centers of excellence and using machine-learning ƒƒ undershooting the potential—some banks just
techniques. They and many others have spent put a technical infrastructure in place and hire
hundreds of millions on their data (especially risk some data scientists, and then execute analytics
data) and on compliance. For them, advanced on a project-by-project basis
analytics is becoming a reflex action, with
commensurate rewards of about €300 million in ƒƒ not asking the right questions, so algorithms
additional annual profit, on average. don’t deliver actionable insights

Most banks we surveyed, however, are struggling. Making it happen

A good many are “started but stuck”: they have Avoiding the pitfalls and accessing the broad
invested significantly in data infrastructure set of opportunities requires CEO leadership
(mostly as a result of regulation) and experimented as banks develop two assets: a strategy for the

5
 transformation and a robust analytics organization Readers may notice something that’s missing from
to assist and empower the businesses as they learn this list: setting the aspiration. That’s because we
to use analytics in their everyday work. (New think every institution, unless its circumstances
technologies and tools are also necessary; see are extraordinary, should set the same aspiration:
sidebar, “A quick take on the CIO agenda.”) to establish analytics as a business discipline—the

A quick take on the CIO agenda

As noted, analytics does not necessarily require a ƒƒ Google-like search engines provide quick and
big investment in IT. But banks must provide the easy access to all of a bank’s data, including
technologies and tools that businesses need to the source, definition, and all other information
access an immense set of high-quality data in real needed to use the data effectively.
time. That data must be well managed and always
available. These technologies include the following: ƒƒ Modern data-exploration and -visualization
tools are helpful to bridge the gap between
ƒƒ Data lakes to scale up and industrialize the advanced business users identifying potential
development and delivery of use cases. There use cases, anomalies, or patterns in the data,
are good and bad ways to use data lakes; we’ve on the one hand, and data scientists using
seen better results when they are developed advanced analytics to predict and optimize, on
with agile techniques based on use cases. At the other.
all costs, banks should avoid building another
enterprise data warehouse or asking data ƒƒ Tools to analyze text, voice, video, and
lakes to do more than they really can. Data images can help to process new data
lakes will not unburden the banks from proper structures—specifically, unstructured data,
governance and quality-assurance processes, such as voice and text. For example, analyzing
but they do offer an opportunity to get things live-chat data has great business potential for
right from the beginning. retaining customers or for next-best-product-
to-buy analyses.
ƒƒ Machine-learning techniques, which have
proved superior in many use cases—the industry ƒƒ Capabilities to leverage real-time data are
will inevitably move toward machine learning important. Deploying analytical models in live
supported by open-source languages. Machine- streams can help embed analytical advice
learning tools are already available in most banks and predictions into customer dialogues. For
but are not used to solve very specific problems. example, most banks have already added chat
(If they are not available, there is a large market capabilities to call centers. Now, banks can
of rapidly moving open-source technologies and add analytics-driven recommendations to the
commercial software products.) Banks need to APIs that support chat to insert a “next best”
keep their options open to the vast innovations suggestion to customers.
taking place in machine learning.

6
go-to tool for the thousands of decision makers value the solution can create. Do not launch
across the bank. Earlier we mentioned analytics as yourself into analysis for the sake of analysis or
a reflex. To extend the metaphor, analytics should into intellectually interesting problems whose
resemble the human nervous system; every part solutions are not actionable.
of the body knows what to do when presented with
certain stimuli. The big difference among banks ƒƒ The smallest edge can make the biggest
will probably be the pace at which they can build difference. Advanced analytics is not about
and train their systems. Just as some parts of your solving your biggest problems; it’s about
brain are trained and some are not, banks will find solving hundreds of small ones that all add up.
that some nerve paths are already working well, Especially in operations, these techniques can
while others must be laid down and taught how help to redefine processes and shorten them by
to react. several steps.

There’s something else missing from our list of ƒƒ Insights live at the boundaries between data
required assets: an additional $100 million or so sets. Remember the bank that combined
of spending. Many bank leaders look at analytics six or seven discrete data sets to build a tool
and fear an outsized investment. That’s not predicting the next product to buy? It realized
unreasonable, since in recent years institutions that lots of relationships become apparent only
have had to spend billions on things they could not when it compared widely varying parameters.
have anticipated, and budgets are very thin. But Banks have massive amounts of data scattered
analytics is not a bet-the-bank investment with no through different departments. Pilots to bring
graceful exits; it’s a short-cycle flow of investments together small samples of information can
with lots of options to kill unsuccessful pilots. The reveal the potential.
small but immediate payoffs from the initial work
can finance the next wave of projects, which in turn ƒƒ Loops beat lines every time. Following
finance more and larger efforts. Once the system a process can be a slog, and to do so you
is built, the investment is over and the margins sometimes have to put on blinders. But that
become enormous—like those of software or mentality is exactly wrong for advanced
tech companies. analytics. Banks that use feedback loops can
not only be faster to market than competitors
A clear strategy centered on high-priority but also arrive there with better products.
applications Ultimately, machines learn just as we do: by
Three elements are essential to the strategy. trial and error.
First, banks need an analytics-ready mind-set.
Analytics transforms everyday work in surprising ƒƒ Design matters. You want people to use your
ways, so leaders must open their minds to the new tools. A beautiful algorithm deserves
possibilities. Our core beliefs about advanced an attractive package that catches the eye of
analytics can help.2 your users. Most of them can’t read code or
understand the output of a model. To act on
ƒƒ Great analytics starts with high-value questions, these insights, they need easily readied and
not data. To guide the discovery process, ask used dashboards that help them make decisions
what problem you want to solve and how much and test potential scenarios.

7
 CDP 2016
Analytics in banking Time to realize the value
Exhibit 2 of 2

Exhibit 2 Ensuring impact has five prerequisites.

An ordered approach to data-analytic enlightenment

1
Study sources
2
Build a data
3
Model your
4
Turn insights
5
Adopt
of value ecosystem insights into actions what works

• Business needs • Internal • Data • Process • Capability

• External modeling redesign building
• Heuristic • Tech and rollout
insights enablement • Change
management

ƒƒ Analytics isn’t enough; adoption is essential. A second element of the strategy is a set of
Use whatever means necessary—incentives, prioritized use cases and a mechanism to create
role modeling, communication, more a pipeline of them. The scope for analytics is vast.
communication—to get decision makers to Anywhere a bank uses rules of thumb or something
use the new tools. Way too often, best-in-class is done “the way we’ve always done it,” analytics
algorithms sit idle in computers because users can probably make improvements. The CEO must
do not trust what they regard as a black box, fear lead the hunt for these issues and help prioritize
the impact it could have on their roles, or simply them. Critically, at the beginning, the chosen use
do not want to go through the discomfort of cases should not be limited to applications in which
change. Creating analytics is like putting jet fuel analytics could produce a substantial uptick in
in your car. At the end of the day, if the driver results; they should also include areas where scale
does not develop the skills needed to drive can be increased quickly, to avoid the “pilot trap.”
faster, the effort is wasted.
Most of the potential use cases are relevant to
ƒƒ Analytics is a team sport. The skills banks need every banking business. They include commercial
to make analytics work cannot be contained applications: cross-selling and upselling, customer
within a single person—at least not yet. Your acquisition, reducing churn, and winning back
teams must include true experts on data science, customers. Business-improvement levers (such as
engineering, data architecture, and design. dynamic and value pricing, credit underwriting,
Faking it with people who do a little bit of sales-area planning, yield and claims management,
everything won’t work. fraud detection, call-center routing, and workforce

8
planning) are also relevant for most banks. While classical steps of successful change management
the first couple of use cases can be introduced will be essential: role modeling the new behavior,
top-down or outside-in, it is just as important clearly explaining why change is needed, building
to encourage everybody in the bank to become the skills of the businesses so they can succeed
creative and make suggestions—while always with the new tools, and reinforcing the bank’s
ensuring a clear path to creating value. To avoid commitment through formal mechanisms (such
discouragement, long validation and delivery as incentives).3
cycles need to be shortened. Innovation labs can
help accelerate the process. A powerful analytics network
The businesses will need help to design analytics
Finally, a strategy should set out a vision for how systems, to build applications exploiting them, and
the bank will use analytics applications. For each to promote adoption. Banks will want to establish
use case the bank is considering, it should start a central team that supports these needs. But the
by asking what problem holds back the business last thing they should do is build another silo. What
from having a greater impact. It can then work they require is a networked structure, a kind of
through a set of five steps: identifying the source nervous system. More than anything, banks must
of value, considering the available data (easier to have open channels and accessibility to make a
do with a data lake, as we describe in the sidebar), knowledge of analytics pass freely throughout the
identifying the analytics technique that will enterprise. An analytics center of excellence, the
respond to the problem and probably produce spine of such a system, will probably need some or
insights, considering how to integrate analytics all of the following components:
into the workflow of the business, and anticipating
the problems of adoption (Exhibit 2). ƒƒ New roles and responsibilities in data
management, advanced analytics, and
Will sellers use the tools? If not, why not? What are campaign execution. Perhaps the most critical
their needs, and how can you make the analytical role is that of a chief data officer responsible not
tools responsive to them? Sometimes the answer only for analytics strategy and its integration
involves bundling insights from algorithms with into business units but also for defining data
useful data for sales managers in an app that they management’s roles and responsibilities,
can use on external visits. Other times, a bank will monitoring the quality of data, and ensuring
have to change the way it develops campaigns and regulatory compliance.
pushes them to the front line and to customers.
Finally, in many other cases the bank will have to ƒƒ Imaginative channels, such as innovation labs,
develop a group of high-performing champions who to bring analytics closer to users.
embrace this discipline and act as role models.
ƒƒ A deep pipeline of analytics talent, cited as a
Embedding analytics in the organization is not top priority by 60 percent of the banks in our
simply a matter of getting specific teams to use recent survey. Banks should typically start
specific tools, though that’s essential. The CEO and with small teams of data scientists, who can
the top team must do much more to communicate work with external partners to absorb the
clearly that analytics is important to the bank necessary competences and skills, and then
and empower everyone to join the revolution. The scale up gradually.

Analytics in banking: Time to realize the value 9

 ƒƒ A clear governance plan with a strong center of They partner with firms that specialize in analytics
excellence (headed by the chief data officer) and and have committed themselves to making
well-defined responsibilities for the governance strategic investments to bolster their analytics
and quality of data and for the execution of capabilities. Within a couple of years, these leaders
analytics within the businesses. may be able develop a critical advantage. Where
they go, others must follow—and the sooner the
ƒƒ “Gold standard” data-management processes better because success will come, more than
that define clear accountability, maintain the anything else, from real-world experience.
quality of data, and manage metadata, the data
life cycle, and controls. 1 We interviewed executives at 13 global and regional banks

based in ten countries across Europe and the Middle East.

ƒƒ Bulletproof data-quality controls, starting at 2 For more, see Helen Mayhew, Tamim Saleh, and Simon

the very beginning of the data life cycle. They Williams, “Making data analytics work for you—instead of
should be automated wherever possible and the other way around,” McKinsey Quarterly, October 2016,
McKinsey.com.
monitored by a set of key quality indicators, 3 See Tessa Basford and Bill Schaninger, “The four building blocks
so that the businesses are accountable for of change,” McKinsey Quarterly, April 2016, McKinsey.com.
collecting the right data.
Amit Garg is a partner in McKinsey’s New York office,
Davide Grande is a partner in the Milan office, Gloria
Macias-Lizaso Miranda is a partner in the Madrid
office, and Christoph Sporleder is a senior expert in
More than 90 percent of the top 50 banks around
the Frankfurt office, where Eckart Windhagen is a
the world are using advanced analytics. Most senior partner.
are having one-off successes but can’t scale up.
Nonetheless, some leaders are emerging. Such
banks invest in talent through graduate programs.

10
April 2017
Designed by Global Editorial Services
Copyright © McKinsey & Company